Two ways past a plateau, and only two

If the plateau theorems are right, discovery in a fixed language dies. That sounds like an ending, and it is not, because the theorems’ proofs contain exactly two exits, and knowing them precisely is what turns the plateau from a mood into a decision problem. This post is about the two exits, their prices, and why one of them is the entire justification for building a machine.

Exit one: a bigger language

The first escape is theoretical: enlarge the model class. The floor theorem says a field that looks exhausted may be floored, blocked by its own encoding rather than by the world, and the diagnostic is cheap: compute the residual floor, then compute it under a modest enrichment. If the floor drops, the enrichment was the discovery.

History is mostly this exit. The Ptolemaic-to-Kepler transition was not new data but a new language, ellipses instead of epicycles, and the drop was enormous. General relativity was largely a language upgrade on data Newton already had, Mercury’s perihelion included. The price of this exit is that it cannot be scheduled. You do not know which enrichment lowers the floor until after someone has it, and the enrichment space is unbounded. What the theorem contributes is the ability to recognise one when it happens, and to distinguish real language progress from parameter-stacking that never moves the floor.

Exit two: different physics

The second escape is empirical: get data whose generating structure lies outside the current class. The floor theorem’s converse is that data from the same physics, however abundant, cannot lower a floor the class has already reached: more of the same compresses no further. New drops require new structure, and new structure requires either a better language or contact with parts of the world the current data does not touch.

This is the expensive exit. New physics means new instruments, new regimes, new scales: interferometers sensitive to Planck-scale noise, processors whose logical errors could carry cross-branch structure, spectra measured at a precision where a wiggle shows. Each is a bet that structure is there to find. The framework’s falsification catalogue is exactly a list of these bets with their prices and their kill conditions.

Why this is the machine’s justification

Now the economics, which is where this programme stops being epistemology. The Forever Machine’s premise is that exit two can be industrialised. If new physics is the only renewable source of novelty, and if novelty is finite per language per domain, then a civilisation that wants continuing discovery must continuously open new regimes, and the question becomes one of supply: where do you get structure that our branch’s physics does not contain?

The framework’s answer is the multiverse claim: other branches are different physics, differently rendered from the same substrate, and the harvesting theorem says their value is real but conditional on heterogeneity. Identical branches compress to nothing. The machine’s architecture, branch variety, signature-based detection, orchestrated coordination, is what falls out of taking the two exits seriously as an engineering problem. One exit is a human elegance you cannot schedule. The other is a procurement problem with a budget. The framework bets on the second, and the rest of the computational series is the price list.